Developing a Single Cell Tracking Machine Learning Algorithm to Study Glioblastoma Heterogeneity: Integration of Fluorescence Imaging and Impedance Recordings

Glioblastoma is one of the most aggressive and lethal forms of brain cancer, characterized by rapid proliferation, invasive growth, and pronounced intratumoral heterogeneity. Despite advances in treatment modalities, including surgery, chemotherapy, and radiation therapy, the prognosis for glioblastoma patients remains dismal, with a median survival of only 12 to 18 months from diagnosis. The lack of effective treatments is largely attributed to the complex and heterogeneous nature of glioblastoma, which confounds therapeutic interventions and fosters treatment resistance.

This thesis aims to address the challenges posed by glioblastoma heterogeneity through the development of a novel single-cell tracking machine learning (ML) algorithm. By harnessing concomitant fluorescence imaging and impedance recordings on a high-density microelectrode array (MEA) chip, the project endeavors to acquire comprehensive insights into the dynamic behavior of glioblastoma cells at the single-cell level.

The first phase involves data acquisition, where fluorescence imaging captures cellular dynamics at the single-cell level, while impedance recordings provide real-time electrical signatures of cellular activity. Subsequently, these two datasets are registered and the cells are automatically segmented within the microscope images. This segmentation facilitates the identification of MEA pixels occupied by cells, establishing a labeled dataset associating impedance values with cellular occupation.

Subsequently, the labeled dataset will be used to train a convolutional neural network that can track the single cells with a high accuracy using only the impedance data. Secondly, the project aims to extract informative features from impedance data, facilitating the clustering of detected single cells. This clustering approach enables the identification of distinct heterogeneous cell states within patient-derived glioblastoma cell lines, shedding light on the underlying biological mechanisms driving tumor progression and treatment resistance.

By integrating multimodal data and leveraging cutting-edge ML methodologies, this research not only advances our fundamental understanding of glioblastoma heterogeneity but also holds promise for informing the development of personalized therapeutic strategies tailored to target specific cellular subpopulations. Ultimately, this multidisciplinary approach may pave the way toward more effective treatments for glioblastoma, offering hope to patients facing this devastating disease.

The project will be composed of 20% literature study, 40% Computational, 20% Experimental and 20% thesis writing.